K100 quicker than a K400 !!

[Russ Hancock_1]

Can anyone help?
Put basically, we have a K400 that's running much slower than out existing K100, performing simple 'tricks' like copying /stand to /tmp.
(for example 8sec on K100 and 24sec on the K400) at a time of very little loading?!

the K400 has 4*100Mhz processors, 1gb ram, the K100 has 1*1Mhz processors and 512Mb ram.

As you guys seem to understand the SAR results, I've attached some incase that helps.

Cheers
Russ.

[harry d brown jr]

Do both have identical disk systems? Are they using the same IO cards? Are they in the same slots? Are both of them up to date on patches?


live free or die
harry

[Marco Paganini]

Hello Russ,

What about the disk/controllers models/configurations? And installed software?

Please send the output of:
- ioscan -funCdisk
- swlist -l product

Regards,
Paga

[A. Clay Stephenson]

Hi Russ:

On these tasks, 4 CPU's vs. 1 is almost meaningless. It's only when you put some load on the box that multiple CPU's will help at all. I suspect differences in tuning between the 2 boxes and also that the disk types are different. If you will do a sysdef of both boxes or a kmtune (if 11x) that will probably help.

[Sanjay_6]

Hi Russ,

Take a look at the thread below,

http://us-support.external.hp.com/cki/bin/doc.pl/sid=c8e3e5bd0a556ac3bb/screen=ckiDisplayDocument?docId=200000057441905

Hope this helps.

Regds

[Roger Baptiste]

<>

What sort of DISKS the systems have??
Are they identical?? IS this time consistently the same??



More processors doesn't necessarily mean superiority for all cases. It applies when there is CPU hogging processes. Simple copies do not fall under that category. Same applies to RAM. So, i wouldn't look too much into processor numbers or RAM size for copy commands , when the system is otherwise idle.

Can you post your ioscan -nfCdisk output?
and also the ioscan -nfCext_bus ?

HTH
raj

[Santosh Nair_1]

It looks like the system spends a lot of time waiting for IO as noted by the %wio, this means that the CPU is not doing anything but waiting form something from the IO subsystem, i.e. typically disks. Also looking further at the output, it looks like the disk at c3t8d0 is being hit hard. You might want to move some of the filesystem on that disk to some other disks in order to distribute the IO. Also, consider getting another controller to spread the IO.

-Santosh

[Russ Hancock_1]

Thanks for the response guys.
I guess a bit more info is required. (see attached).

Both machines are pretty similar, both with 4 internal disks hanging off a FastWide SCSI and an additional 8 disks in a storage array running from a Wide SCSI card, some LVM but most just whole disk mounts, NFS.

I understand that multiprocessors and greater memory won't have a great affect the speed of a simple file copy, but I was expecting it to a similar duration between the 2 machines, not 3-4 times slower!! hence my concern.
Thanks.

[A. Clay Stephenson · Accepted solution]

Hi Russ:

You have 2 problems. My psychic powers wqere working. You applied one of those tuned parameter sets. It set the timeslice to 1. Bad boy!! Set it to 10. Also, dbc_max_pct is 2%; that's only 20MB in your case. I suggest that you forget forget dynamic buffer cache and set bufpages=25600. That will give you a fixed buffer cache of 100MB - a fairly good starting value in your case.

Regards, Clay

[Russ Hancock_1]

Hi Santosh,

Don't say that, with just installed a new SCSI card in this machine, as it turned out the original configuration had all 12 disks hanging off a FastWide!!
As for the disk you mentioned that contains one of the oracle tablefiles - only. but that's probably why it's quite loaded.

Cheers
Russ.

[A. Clay Stephenson]

Hi again Russ:

You didn't respond but I want to emphasize that your timeslice value of 1 is your fundamental problem. With that value, you have a certified dog regardless of application and regardless of I/O configuration.

Regards, Clay

[Russ Hancock_1]

Thanks for your help guys, I'll try the timeslice as soon as I can get every one off the system.... Plenty of reading material suggested so I'll go through that as well..
Cheers
Russ.

[Russ Hancock_1]

One last question..
Is there a recommended method of testing a machines performance, regardless of the applications running on it? Obviously a simple copy doesn't test the system in the right way, have any of you found a more reliable test?
Cheers
Russ.

[Roger Baptiste]

Hi,

<>

For that you would need benchmarking software tools! I normally do it only for the
I/O subsystem, since that's where most of the problems lie. For that, you can use the I/O test tool from soliddata . Link is
http://www.soliddata.com/products/iotest.html

From the CPU end, we normally go by the specs of the vendor on how it should perform and what sort of load it can take. IN case the application is loading more cpu, we add resources to the box as per the needs demands. Same applies for memory subsystem.

When you get the system rolling for the users, make sure the kernel parms are configured the way it should be: Refer

http://docs.hp.com//hpux/onlinedocs/os/KCparams.OverviewAll.html

THis would need tuning based on what applications you are running on the box.

HTH
raj